[erlang-questions] How does gproc register global names?

[Ulf Wiger]

On 26 Oct 2010, at 18:49, Pablo Platt wrote:

> Why doesn't mnesia have a 'light weight' transaction that works like gproc?

Mnesia has a slightly different charter, ensuring consistent persistency.


> When a new leader is elected, how does other nodes knows which names it doesn't know about?

In the Mod:elected() callback, the newly elected leader can 
pass data to the others, which they receive in Mod:surrender().

> I'm using gproc to register gen_server processes.
> It might be useful to have gproc:cast and gproc:call that use gen_server:cast and gen_server:call in addition to gproc:send.

True, but my idea was that OTP integrate gproc and make gen.erl
aware of gproc-registered names. :)

> Is it possible to count all registered processes without using a counter?

It would be possible if the function gproc:pattern/2 were exported, or 
a function gproc:select_count/2 were added, similar to select/2 et all,
which are already there.

> Is it possible to page thorough all registered processes?

Yes, with either QLC (using gproc:table/1) or using gproc:select/3 (using 
the Limit argument to fetch a give number of objects at a time.

BR,
Ulf W


> 
> Thanks
> 
> 
> 
> From: Ulf Wiger <ulf.wiger@REDACTED>
> To: Pablo Platt <pablo.platt@REDACTED>
> Cc: erlang-questions@REDACTED
> Sent: Tue, October 26, 2010 5:31:24 PM
> Subject: Re: [erlang-questions] How does gproc register global names?
> 
> 
> It is correct that gproc uses gen_leader.
> All registration reguests are passed on to the leader process,
> and it broadcasts the results to the other candidates & workers.
> Monitoring of registered processes is done by the local gproc 
> instance.
> 
> When a new leader is elected, it sends its list of registered names
> to the others; if they happen to have names that the new leader 
> doesn't know about, they will send them to the new leader.
> 
> This is not terribly unlike how global does things, although it uses
> a multicall mechanism to handle atomicity. In measurements that 
> I've made, the gproc method is much more efficient.
> 
> OTOH, global has facilities to handle partitioned networks. Gproc
> doesn't... yet.
> 
> Using mnesia would not give better consistency, unless transactions
> are used - but that would kill performance. And there are no persistency
> reasons to use it either. Gproc uses ets tables, which are rigged to survive
> even if the gproc process dies; the restarted gproc scans the table and 
> sets new monitors.
> 
> BR,
> Ulf W
> 
> On 26 Oct 2010, at 15:48, Pablo Platt wrote:
> 
> > Hi
> > 
> > I'm using gproc to register local names and properties.
> > I've read that gen_leader is needed to use gproc in the global scope.
> > 
> > How does the data get synced across all nodes?
> > Why isn't mnesia being used for that?
> > How do you solve race conditions and failures?
> > 
> > Thanks
> > 
> > 
> > 
> 
> Ulf Wiger, CTO, Erlang Solutions, Ltd.
> http://erlang-solutions.com
> 
> 
> 
> 
> ________________________________________________________________
> erlang-questions (at) erlang.org mailing list.
> See http://www.erlang.org/faq.html
> To unsubscribe; mailto:erlang-questions-unsubscribe@REDACTED
> 
> 
> 

Ulf Wiger, CTO, Erlang Solutions, Ltd.
http://erlang-solutions.com